Available online at www.sciencedirect.com ScienceDirect

Explorations in Economic History 58 (2015) 1–21 www.elsevier.com/locate/eeh

Urbanization without growth in historical perspective☆

Remi Jedwaba,⁎, Dietrich Vollrathb

a Department of Economics, George Washington University, 2115 G Street, NW, Washington, DC 20052, USA b Department of Economics, University of , 201C McElhinney Hall, Houston, TX 77204, USA

Available online 11 September 2015

Abstract

The world is becoming more and more urbanized at every income level, and there has been a dramatic increase in the number of mega- in the developing world. This has led scholars to believe that development and are not always correlated, either across space or over time. In this paper, we use historical data at both the level and level over the five centuries between 1500–2010 to revisit the topic of “urbanization without growth” (Fay and Opal, 2000). In particular, we first establish that, although urbanization and income remain highly correlated within any given year, urbanization is 25–30 percentage points higher in 2010 than in 1500 at every level of income per capita. Second, while historically this shift in urbanization rates was more noticeable at the upper tail of the income distribution, i.e. for richer , it is now particularly visible at the lower tail, i.e. for poorer countries. Third, these patterns suggest that different factors may have explained the shift in different periods of time. We use the discussion of these factors as an opportunity to provide a survey of the literature and summarize our knowledge of what drives the urbanization process over time. © 2015 Elsevier Inc. All rights reserved.

JEL classification: N9; R1; O1 Keywords: Urbanization without growth; Economic development; ; Urban poverty; Urbanization

Contents

1. Introduction ...... 2 2. Urbanization and development in the data ...... 3 2.1. Urbanization and GDP per capita ...... 3 2.2. Measurement issues and robustness checks ...... 5 2.2.1. Earlier years ...... 5 2.2.2. Later years ...... 6 2.2.3. Comparisons across time ...... 7

☆ We would like to thank Hans-Joachim Voth for encouraging us to write this paper, and for his valuable comments. We are also grateful for seminar audiences at George Washington and George Mason for very helpful comments. We thank the Institute for International Economic Policy at George Washington University for financial assistance. ⁎ Corresponding author. E-mail addresses: [email protected] (R. Jedwab), [email protected] (D. Vollrath). http://dx.doi.org/10.1016/j.eeh.2015.09.002 0014-4983/© 2015 Elsevier Inc. All rights reserved. 2 R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21

2.3. The evolution of mega-cities ...... 7 2.4. Summary of empirical analysis...... 13 3. Urbanization and development in theory ...... 13 3.1. Conceptual framework ...... 13 3.2. Urbanization with growth ...... 14 3.2.1. Agricultural revolutions ...... 14 3.2.2. Industrial revolutions ...... 14 3.2.3. Resource revolutions ...... 14 3.2.4. Agglomeration effects...... 15 3.3. Urbanization without growth...... 16 3.3.1. Urban technologies ...... 16 3.3.2. Urban amenities and preferences ...... 16 3.3.3. Urban bias ...... 16 3.3.4. Rural poverty ...... 17 3.3.5. Internal urban growth ...... 17 4. Discussion and conclusion ...... 18 Appendix A. Data Sources...... 18 Appendix B. Supplementary data ...... 19 References ...... 19

1. Introduction In this paper we examine the relationship of urbanization and development levels from 1500 to Urbanization is often considered a hallmark of 2010, with the goal of discovering whether in fact this economic development (Kuznets, 1968; Bairoch, 1988). relationship has changed appreciably in the late 20th Urbanization rates and city sizes have often been used as century. We first establish that the world is becoming empirical proxies for the level of income per capita (De et more urbanized at every level of income per capita over al., 1993; Acemoglu et al., 2002; Dittmar, 2011). At a time, with the shift on the order of 25–30 percentage global level they have trended upward together from the points for all countries. However, this occurred in Neolithic Revolution, when an appreciable agricultural different periods of time for different parts of the income surplus first allowed for towns and cities to form, to the distribution. From 1500 until the mid-20th century we current day when roughly half of all people live in urban find that urbanization rates rose dramatically for the areas (, 2014). richest countries, but remained roughly constant (i.e. Recently Fay and Opal (2000) documented what they close to zero) for countries at the lowest levels of income called urbanization without growth in developing coun- per capita. However, from the mid-20th century until tries of the late 20th century. In a similar vein, Glaeser today, urbanization has occurred primarily among the (2013) shows that there is “poor country urbanization” at poorest countries in the income distribution. In this sense, the bottom of the income distribution, and Jedwab and urbanization without growth is not unique to developing Vollrath (2015) document that mega-cities are increas- countries in the 20th century. Rather, it appears to have ingly located in poorer countries. These papers suggest been occurring for several centuries. the possibility that something fundamental has changed A consequence of this difference in the timing of in the relationship of urbanization and development in the urbanization is that the location of “mega-cities”–the late 20th century compared to prior experience.1 largest urban agglomerations in a given time period – has changed demonstrably over time. From 1500 to the mid-20th century the largest mega-cities in the world 1 The theme of urbanization without growth is not entirely new to economics and . (Hoselitz, 1953)explainedthaturbanization, tended to be in countries that were relatively rich and industrialization, and economic development are not always correlated. developing rapidly (e.g. and New York). From (Hoselitz, 1955)suggestedthatmanycitiesinthedevelopingworldwere that point until today increased urbanization rates, “parasitic” as opposed to “generative”.Similarly,(Bairoch, 1988)discusses combined with large bases, mean that the the processes of “over-” or “hyper-” urbanization in the process of largest mega-cities now tend to be located in poor development. Lastly, (Davis, 2007)studythedramaticexpansionofslums and the rise of an “informal urban proletariat” in the developing world. countries (e.g. and ). We document below Davis thus also argues that urbanization has been disconnected from how the location of mega-cities has changed over time economic growth. along with the urbanization rate. R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21 3

Last, we review the literature on urbanization and 2. Urbanization and development in the data development for explanations of why this disconnect in the urbanization process occurred, and why it occurred 2.1. Urbanization and GDP per capita starting in the mid-20th century. We note that there are both positive and negative types of urbanization without Our first crude examination of the urbanization/ growth. Technologies such as clean water, sewers, public development relationship is in Fig. 1. This plots transportation and multi-story buildings increase the urbanization rates against log GDP per capita for a total relative value of living in urban areas and generate larger of 1319 observations that we collect. Data sources are cities, even if they may have no direct impact on available in the Appendix A. The positive correlation is productivity or output per worker. These positive forces clear and quite strong. Log GDP per capita explains about seem to predominate in the earlier era leading up to the two-thirds of the variation in urbanization rates. The slope mid-20th century. On the other hand, urban bias in estimate indicates that tripling GDP per capita (equivalent policies or rural poverty represents negative urbanization to raising log GDP per capita by 1) raises the urbanization that appears to lower welfare even if it has no immediate rate by about 20 percentage points. This represents an impact on output per worker. While there certainly may equilibrium relationship, and not a causal one. Economic be positive forces driving urbanization in the late 20th development both drives urbanization and is driven by century in developing countries, their experience appears urbanization, a subject we will return to in the next to be explained more by the negative sources of section of the paper. But this figure shows how tight that urbanization without growth. equilibrium relationship is across the entire sample. To continue, we first review data on urbanization and If the equilibrium relationship between development development levels as well as data on the size and living levels and urbanization was unchanging, then what we standards of mega-cities over time. Once we have would see across time is that countries would “climb up” established the patterns in this data, we review theories the line plotted in Fig. 1. As countries began the of urbanization and growth and how they can be used to development process this would extend the distribution interpret those patterns. of points up the line, but the line itself would remain stable over time. However, this is not what we see when we examine different time periods. Rather than a stable relationship, the link between urbanization and development has evolved over time. Fig. 2 plots the data from three years: 1500, 1950, and 2010. The relationship between urbanization and development is estimated and plotted separately for each of the three years. As can be seen, there are both level and slope differences across the years.

Fig. 1. Urbanization and economic development, 1500–2010. Notes: We use data for 2217 observations belonging to 159 countries for the following years: 1500 (24), 1700 (18), 1800 (25), 1850 (23), 1870 (46), 1910 (70), 1950 (159), 1970 (159), 1980 (159), 1990 (159), 2000 (159), and 2010 (159). The main sources for the urbanization rate (%) are Bairoch (1988); Acemoglu et al. (2002); Malanima and Volckart (2007); United Nations (2014); Jedwab and Moradi Fig. 2. Urbanization and economic development across time: 1500, 1950 (Forthcoming). We use Maddison (2008); Bolt et al. (2014) to obtain and 2010. Notes: We use data for 342 observations belonging to 159 log per capita GDP (PPP, constant 1990 dollars). Together, the 159 countries for the following years: 1500 (24), 1950 (159) and 2010 (159). countries account for 99% of the in 2010. See the notes of Fig. 1 for the list of sources used to reconstruct the data. 4 R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21

among the richest nations had only increased slightly by 2010. At GDP per capita of $8,100 (9.0 in logs) urbanization rates in 2010 were still around 60–80%, similar to their level in 1950. Thus the slope of the income/urbanization relation- ship, while still positive, had a lower absolute value in 2010 than in 1950. It is now roughly the same as in 1500, with a tripling of GDP per capita associated with only a 13 percentage point increase in urbanization rates. Over the very long run, then, the urbanization rate has risen across the entire spectrum of development. The minimum level of urbanization appears to have risen from zero in 1500 to 20% in 2010.2 Fig. 3. Yearly correlation between urbanization and economic The changing relationship between urbanization and development, slope and constant, 1500–2010. Notes: For each year, development over time can be most easily seen in Fig. 3. we regress the urbanization rate (%) on log per capita GDP (PPP, For each separate year, we regressed the urbanization rate constant $1990) and show the coefficient and the constant. The dashed vertical line represents the year 1910. See the notes of Fig. 1 for the against log GDP per capita. Against the left-axis is plotted list of sources used to reconstruct the data. the coefficient on log GDP per capita. From 1500 to 1910 the slope of the relationship tends to move upwards, although as one can see there is substantial variation from In 1500, development is low and related to urbanization, one year to the next, in part due to the limited sample sizes but the strength of that relationship is not as strong as in we have to work with for these years.3 the overall sample. At log GDP per capita of 6 – roughly However, there is clearly a distinct change after $400 in 1990 dollars, equivalent to Burundi in 2010 – the 1910. Here the relationship has a slope of 20, and urbanization is close to zero. But even the most advanced remains there through 1960. The intercept terms of the countries in the 1500 data, Italy and the Netherlands both regressions are also plotted in Fig. 3 against the right with roughly $1,800 per capita (7.5 in logs), had axis. The pattern seen in these tell us about where urbanization rates of only about 20%. In 1500 tripling exactly the slope changes were taking place over time. GDP per capita was only associated with a difference in In 1500 the constant term is relatively large, and despite urbanization rates of about 12 percentage points. the variation it drifts lower until dropping substantially By 1950, this relationship has changed substantially, in 1910. The drop in the constant along with the higher with a higher slope. In this year, we have that tripling slope in 1910–1950 indicates that urbanization rates GDP per capita is associated with a 20 percentage point were becoming larger for richer countries between 1500 higher urbanization rate. This change is driven by high and the mid-20th century. urbanization rates among the most developed nations. By 1950, a country with GDP per capita of about $1,800 (7.5 in logs) – , for example – was more urbanized than its peers in 1500, reaching a 40% rate on average. At the 2 highest end, with log GDP per capita of 9 (e.g. the United Web Appendix Figs. 1 (1500 to 1950) and 2 (1950 to 2010) show the estimated relationship between urbanization and log income per States with $8,100 per capita) urbanization rates were on capita for selected years in our sample in the indicated interval. The the order of 60–80%. However, it was still the case that figures confirm that urbanization rates have been relatively increasing the poorest countries in 1950 (i.e. those with log GDP per ceteris paribus for the richest countries between 1500 and 1950 and capita of about 6, or $400) had urbanization rates close to for the poorest countries between 1950 and 2010. An alternative way zero. Hence the slope of urbanization with respect to GDP of seeing the evolution of urbanization over time is to look at average urbanization relative to average log GDP per capita for the world in a per capita was quite large. given year. Web Appendix Fig. 3 plots this ratio, which is clearly But if we continue the analysis through 2010, we see rising over time, indicating higher urbanization at the average of log another significant shift of the relationship. In the period GDP per capita over time. after World War II it was the poorest countries that saw 3 Figures showing the standard errors of estimates are available in urbanization rates increase most rapidly. At GDP per the Web Appendix Fig. A.4. Given that we have only 18–25 observations in years prior to 1870, these confidence intervals are capita of $400 (6.0 in logs), the urbanization rate was on relatively large, narrowing as the sample roughly doubles in size in the order of 25% in 2010, whereas it had been close to 1870 (46 countries), triples in 1910 (70) and increases sixfold in zero in 1950. At the same time, urbanization rates 1950–2010 (159). R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21 5

After 1970, the slope drops, reaching 13 by 2010. Although urbanization remains an indicator of GDP per capita, this is becoming less reliable as we enter the 21st century. At the same time, the intercept term is becoming larger from 1970 to 2010. This indicates that urbanization rates are becoming larger for poorer countries since the mid-20th century, as shown in Fig. 2.4 For an alternative way of visualizing where the changes in urbanization have come from over time, Fig. 4 plots the predicted level of urbanization at different levels of GDP per capita for each year. For countries at $400 of GDP per capita, the predicted urbanization rate is essentially zero from 1500 to 1910, Fig. 4. Yearly predicted urbanization rate at different levels of and very close to that level until 1970. By 1980, economic development, 1500–2010. Notes: For each year, we plot the however, even countries with this very low level of predicted level of urbanization at the indicated levels of log GDP per GDP per capita begin to have appreciable urbanization capita. The predictions are based on the regression of urbanization on and by 2010 are predicted to have urbanization rates of log GDP per capita in the specified year. The dashed vertical line represents the year 1910. See the notes of Fig. 1 for the list of sources close to 25%. This level of urbanization is similar to that used to reconstruct the data. seen in countries with $3,000 of GDP per capita in 1500 or 1800. A similar dynamic is seen for countries at $1,100 (e.g. Nepal in 2010) in GDP per capita. Between to 2010. But this shift up in urbanization did not take 1500 and 1910, their urbanization rates were around place uniformly over time. From 1500 to the mid-20th 15%, and then after a small jump by 1950 the century, urbanization was associated with the richest urbanization rate for these countries rises to nearly countries in the distribution. But since the mid-20th 40% in 2010. This is an urbanization rate as high as the century the advance of urbanization has switched to the most urbanized countries in 1500. poorest countries. For countries with either $3,000 (e.g. the in 2010) or $8,100 (e.g. ) in GDP per capita, 2.2. Measurement issues and robustness checks Fig. 4 shows that the shift up in urbanization rates occurred earlier in time. Between 1870 and 1950 there The changes in the relationship over time imply that are distinct jumps in the urbanization rates at these one must be careful about using urbanization rates as a levels of GDP per capita, from 30% to almost 45% at proxy for development. Comparisons across years are $3,000 per capita and from 45% to 65% at $8,000. After problematic. Given the shifts in the constant seen in this acceleration in urbanization rates, there have been Fig. 3, it is quite possible for urbanization rates to rise more modest increases since 1950. without any associated change in log GDP per capita. Figs. 3 and 4 show that overall, urbanization rates have Fig. 4 lays this out most clearly. A country that remains been rising at all levels of GDP per capita over time. at $400 in GDP per capita across time would still be Roughly, urbanization rates are 25–30 percentage points expected to see its urbanization rate rise from 0% in higher in 2010 than in 1500 at any given level of income 1500 to almost 25% by 2010. But even within any given per capita. In this sense, there has been urbanization year, although it is true in our data that urbanization and without growth at every level of development from 1500 development are correlated and patterns are surprisingly consistent over time, one has to be careful.

4 While the cross-country relationship between urbanization and 2.2.1. Earlier years development has been shifting over time, a separate question is whether the within-country relationship is similar to the cross-country Measurement errors are a particular issue for the one. Using our panel, we can perform regressions that include both earliest time periods, precisely the ones in which one country fixed effects as well as year fixed effects. The result of that would hope to use urbanization rates to proxy for regression is a slope estimate of 6.6, significant at 1%. This is half of non-existent GDP per capita data. In years such as the estimated effect in either 1500 (11.3) or 2010 (13.1), and one-third 1500, the sources of data for both urbanization and of the estimated slope in 1950 (19.7). Web Appendix Fig. A.5 then shows the slope and the constant for each year relative to the omitted GDP per capita are subject to much uncertainty, and year 1500 and confirms the patterns seen in the repeated cross- thus the estimated correlation between them may be sectional regressions of Fig. 3. spurious. If measurement errors in GDP per capita are 6 R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21 simply noise, then we will be under-estimating the slope more observations are available and when urbanization of the relationship. On the other hand, if the measure- rates are better documented for all countries. ment errors are endogenous to the urbanization rate, then we are likely over-estimating the slope, which would be however less of an issue for our analysis. 2.2.2. Later years We use the Bolt et al. (2014) update of Maddison Measurement errors are also an issue for the latest (2008) to measure GDP per capita, but there are questions time periods, especially relative to the earliest time regarding the accuracy of Maddison's estimates of GDP periods. We first check that our results are not entirely per capita for not only the deep past, but through the 19th driven by rich countries, in particular since the century (Clark, 2009). The main concern for our purposes continuous development of countries extends the is not whether Maddison gets the level of GDP per capita distribution of points to the right along the x-axis of exactly right, but whether the relative values across per capita income. We find similar patterns if we use countries is accurate. A possible explanation for the only developing countries in any year as defined by the patterns we have seen is that Maddison's estimates are International Monetary Fund (Web Appendix Fig. A.6), systematically biased. The biggest concern is that in or countries that have output per capita below the generating the estimates of GDP per capita in very early median in any year (Web Appendix Fig. A.7). years, Maddison used an assumed relationship between While the inconsistency in thresholds does not appear urbanization rates and development to infer GDP per to be problematic for earlier years, there remains the capita. In this case, our regression for 1500 (or any year question of whether the population counts are accurate for prior to the availability of quality GDP data) is simply later years. This is actually less of an issue for those years reconstructing this assumed relationship. In this case, as cities grew well beyond the 5000 person threshold, there is no evidence that urbanization is a good proxy for making inaccuracies in their population counts immate- development in 1500, despite the fact that we may see a rial to whether a city was larger than 5000 inhabitants. robust relationship in 1910, 1950, or later. However, this Additionally, we use data on the type of urban should only lead to an upward bias. definition used by all countries for most of the 1950– Regarding the urbanization data itself, this also is 2010 period. Among the 159 countries of the post-1950 subject to issues with measurement. There are issues in sample, 81 countries use an explicit threshold to define a defining what constitutes a city, and in making accurate locality as a city, whereas 78 countries use a more estimates of the population of those cities. However, administrative urban definition. Web Appendix Fig. A.8 Bairoch (1988); Malanima and Volckart (2007) both shows the kernel distribution of the threshold used for the use a threshold of 5000 inhabitants to define a city in 81 countries using this type if definition. The median their historical data. While we use several additional threshold is 2500 inhabitants but the mean threshold is sources to supplement their data sets, it is fortunate for 4362, not far from the 5000 threshold used by Bairoch us that many of them also use the standard threshold of (1988); Malanima and Volckart (2007). Controlling for 5000 inhabitants, which should minimize exogenous the type of definition used (via a dummy equal to one if measurement errors. But endogenous measurement the country uses an urban definition based on a threshold, errors could still create biases in the estimated and the threshold itself) or restricting ourselves to relationship between urbanization and development. If, countries using a threshold close to 5000 inhabitants (by for example, it is in particularly poor countries that the restricting our sample to countries with a threshold population in cities of 5000 is underestimated then true between the 10th and 90th percentiles) does not change urbanization rates would be higher, and the estimated the results (see Web Appendix Figs. A.9 and A.10). slope of the relationship lower.5 However, the pattern of Definitional issues regarding urbanization rates have urbanization without growth between 1500 and 1950 is been raised for Sub-Saharan Africa in particular (Potts, particularly visible for the later years of the period 1995, 2012; World Bank, 2009). Web Appendix Fig. (post-1870, see Web Appendix Fig. A.1), when many A.11 shows that if we exclude Sub-Saharan Africa, there is no change to the patterns in the data. Alternatively, we can include fixed-effects (Web Appendix Fig. 5 Xu et al., (2015) provide revised measures of the urbanization rate A.12) or exclude countries that are outliers in general in from 1100 to 1900, and indicate that the rate is roughly 12% (Web Appendix Fig. A.13), and receive similar results. in 1500, rather than the 3.8% that we use from (Bairoch, 1988). If they are correct and errors such as these were replicated across other The six considered are North America, South countries, this would likely reduce the estimated relationship of America, Europe, Africa, Asia and Oceania. We then urbanization and log GDP per capita in 1500. categorize as outliers countries for which the ratio of the R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21 7 urbanization rate to log per capita GDP is below the 10th Finally, results are robust to using either population percentile or above the 90th percentile in 2010. weights (Web Appendix Fig. A.19) or area weights Lastly, an alternative measure of urbanization that is (Web Appendix Fig. A.20) in the regressions. Results independent of the urbanization rate is the total are thus not driven by the larger countries. employment share of industry and services. As shown by Gollin et al. (2013) using census and survey data for 2.2.3. Comparisons across time about 80 countries, agriculture is clearly a rural-based Our baseline results, as well as the robustness checks sector while industry and services are clearly urban-based just described, all operate under the assumption that GDP sectors. Structural change out of agriculture and into is comparable over time. That is, a GDP per capita of industry and services thus reflect the urbanization of both $1,000 in 1500 is presumed to be an equivalent living the economy and the society. Data is rather unfortunately standard to $1,000 in 2010. This assumption is almost not available for the total employment share of these certainly untrue, though, given the differences in products urban sectors before 1980. Yet we find the same patterns available in 1500, 1950, 2010, and years in between of urbanization without growth for the post-1980 period (Nordhaus, 1996; Hausman and Liu, 2014). This issue is when urbanization is proxied by structural change out of analogous to issues in comparing living standards across agriculture (see Web Appendix Fig. A.14). countries with very different consumption baskets While there are measurement issues for urbanization (Deaton and Heston, 2010). GDP per capita may be a rates, there are also measurement issues for countries' poor measure of “true” living standards, even if it does levels of economic development. As GDP is a flow not suffer from measurement error in the sense used in concept, it tends to be more variable than the urbanization our prior robustness checks. rate, which captures a stock concept. For example, many This means that “true” living standards and urbaniza- developing countries experienced a recession in the tion could have a stable relationship over time, but using 1980s–1990s. If there are fixed costs in building cities, GDP per capita mistakenly leads us to infer changes have such as durable housing (Glaeser and Gyourko, 2005), an occurred. Perhaps the increased urbanization rates in poor economic recession will reduce per capita GDP without countries between 1950 and 2010 is associated with necessarily affecting the urbanization rate. This may distinct improvements in living standards due to the create urbanization without growth, but only because introduction of new products (e.g. electricity, mobile income temporarily decreases, and not because urbani- phones), even though measured GDP per capita has not zation permanently increases. changed much. Thus “urbanization without growth in Our results are first robust to excluding country/ GDP per capita” may not imply that there is “urbaniza- years experiencing a recession between two periods tion without growth in living standards”. With that caveat (Web Appendix Fig. 15). There are insufficient in mind, our results show that urbanization is higher at all observations prior to 1800 to make any useful analysis. standards of living as captured by GDP per capita. Results also hold when using a 20-year moving average of GDP (Web Appendix Fig. A.16). We focus on the 2.3. The evolution of mega-cities 1960–2000 period because we cannot build 20-year moving averages for other years. It has been argued that The rise of “mega-cities” over time mirrors the capital could be a better measure of long-run changes in urbanization just documented. By mega- development than per capita GDP (Henderson et al., cities we mean the largest urban agglomerations observed 2013b). We thus test that results are similar when using in a given time period. The absolute size of mega-cities primary school completion in place of GDP (Web has grown over time across countries at all levels of Appendix Fig. A.17), or average years of school in development, similar to the urbanization rate. But there place of GDP (Web Appendix Fig. A.18). However, was significant growth in mega-cities that occurred in we only have data on education from 1960. Lastly, the rapidly industrializing countries of the 19th and the literature has also shown how poorly measured early-20th century before spreading across the rest of per capita GDP is for many developing the developing world in the 20th century. countries, particularly Sub-Saharan Africa (Young, Fig. 5 shows two maps, first for 1900 (Panel A) and 2012; Henderson et al., 2013b; Jerven, 2014). second for 2010 (Panel B). In each, the 100 largest cities However, we have shown in Web Appendix Figs. in that year are plotted by location, with the size of the A.11 and A.12 that results are unchanged when circles indicating absolute size. In 1900, it is clear that excluding Sub-Saharan Africa or including continent the vast majority of these mega-cities are concentrated fixed-effects to do within-continent comparisons. within what would be considered the developed world 8 R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21

Fig. 5. Location of world's 100 largest mega-cities across time. Panel A: 1900 height 18 pt depth 10 pt width 0 pt. Panel B: 2010 height 18 pt depth 10 pt width 0 pt. Notes: Population sizes are in millions. The main sources of the data are Chandler (1987); United Nations (2014). We use the year 1900 instead of the year 1910 as in the analysis on urbanization rates, because Chandler only lists the 100 largest cities for the year 1900. at that point. The largest cities in 1900 are London several large cities in Europe and North America, but by (6.5 million), New York (4.2), and (3.3). Indus- 2010 mega-cities are located across Central and South trial centers such as Manchester (1.4) and Birmingham America, Africa, and Asia. The largest city in 2010 is (1.2) are among the largest 15 cities in the world in (37 million), in a currently developed nation. But 1900. (1.1) and (1.1) are the only cities the next largest are (21.9), City (20.1), over 1 million inhabitants in 1900 that are located in (20.0), and Sao Paulo (19.7), all in what would non-industrializing countries of that period. be considered developing nations. Further, cities such as In contrast, in 2010 (Panel B) the geographic location Dacca (14.7), (11.9), Lagos (10.8), and of mega-cities has shifted away from the richest (9.4) are among the largest cities in the world despite developed nations and towards less-developed ones. being in located in some of the poorest countries. The map in Panel B shows a much wider geographic Table 1 lists the largest 30 cities in selected years spread of the largest cities in the world. There are still from 1500 to 2010, and a few general patterns emerge. Table 1 World's 30 largest megacities (millions), 1500–2010. Rank 1500 1700 1825 1850 1875 1900 1950 2010 1 Beijing 0.7 0.7 Beijing 1.4 London 2.3 London 4.2 London 6.5 New York 12.3 Tokyo 36.8 2 Vijayanagar 0.5 Tokyo 0.7 London 1.3 Beijing 1.6 Paris 2.3 New York 4.2 Tokyo 11.3 Delhi 21.9 .Jda,D olah/Epoain nEooi itr 8(05 1 (2015) 58 History Economic in Explorations / Vollrath D. Jedwab, R. 3 0.4 Beijing 0.7 0.9 Paris 1.3 New York 1.9 Paris 3.3 London 8.4 Mexico 20.1 4 0.3 London 0.6 Paris 0.9 Guangzhou 0.9 Berlin 1.0 Berlin 2.7 Paris 6.3 Shanghai 20.0 5Tabriz0.3Paris0.5Tokyo0.7Istanbul0.8Vienna1.0Chicago1.7Moscow5.4SaoPaulo19.7 6Istanbul0.2Ahmedabad0.4Istanbul0.7Tokyo0.8Istanbul0.9Vienna1.7BuenosAires5.1Osaka19.5 7 Paris 0.2 0.4 St. Petersburg 0.4 New York 0.6 Beijing 0.8 Tokyo 1.5 5.0 19.4 8 Guangzhou 0.2 Isfahan 0.4 Hangzhou 0.4 Mumbai 0.6 Mumbai 0.8 St. Petersburg 1.4 Kolkata 4.5 New York 18.4 9Nanking0.1Kyoto0.4Kyoto0.4St.Petersburg0.5Philadelphia0.8Manchester1.4Shanghai4.3Cairo16.9 10 Cuttack 0.1 Hangzhou 0.3 Naples 0.4 Berlin 0.4 St. Petersburg 0.8 1.4 Osaka 4.1 Beijing 16.2 11 Fez 0.1 Amsterdam 0.2 Osaka 0.4 Hangzhou 0.4 Tokyo 0.8 Birmingham 1.2 4.0 Dacca 14.7 12 Adrianople 0.1 Naples 0.2 0.3 Philadelphia 0.4 Guangzhou 0.7 1.1 Berlin 3.3 Kolkata 14.3 13 Xian 0.1 Guangzhou 0.2 Lucknow 0.3 0.4 Kolkata 0.7 Beijing 1.1 Philadelphia 3.1 14.2 14 Ayutthaya 0.1 Aurangabad 0.2 Vienna 0.3 Liverpool 0.4 Liverpool 0.7 Kolkata 1.1 3.0 14.1 15 0.1 Lisbon 0.2 Xian 0.3 Kolkata 0.4 Glasgow 0.6 Boston 1.1 St. Petersburg 2.9 Istanbul 12.7 16 Suzhou 0.1 Cairo 0.2 Lisbon 0.3 Naples 0.4 Manchester 0.6 Glasgow 1.0 Mexico 2.9 Rio de Janeiro 12.4 17 Venice 0.1 Xian 0.2 Moscow 0.3 Manchester 0.4 Moscow 0.6 Osaka 1.0 Mumbai 2.9 Los Angeles 12.2 18 Naples 0.1 Seoul 0.2 Cairo 0.2 Moscow 0.4 Birmingham 0.5 Liverpool 0.9 Detroit 2.8 Manila 11.9 19 0.1 Dacca 0.2 Berlin 0.2 Glasgow 0.3 Boston 0.5 Istanbul 0.9 Boston 2.6 Moscow 11.5 20 Milan 0.1 Ayutthaya 0.2 0.2 Suzhou 0.3 Naples 0.5 Hamburg 0.9 Cairo 2.5 11.2 21 0.1 Venice 0.1 Patna 0.2 0.3 Cairo 0.4 Buenos Aires 0.8 2.5 Lagos 10.8 22 Fuzhou 0.1 Suzhou 0.1 0.2 Osaka 0.3 Chicago 0.4 Budapest 0.8 Manchester 2.4 Paris 10.5 23 Delhi 0.1 Nanking 0.1 Amsterdam 0.2 Madras 0.3 Hyderabad 0.4 Mumbay 0.8 Sao Paulo 2.3 10.2 24 Gent 0.1 Rome 0.1 Dublin 0.2 Lucknow 0.3 Lyon 0.4 0.8 Birmingham 2.2 Seoul 9.8 25 Kaifeng 0.1 Smyrna 0.1 Seoul 0.2 Birmingham 0.3 Madras 0.4 Rio de Janeiro 0.7 2.1 London 9.7 –

26 Moscow 0.1 Srinagar 0.1 Kolkata 0.2 Xian 0.3 Madrid 0.4 Warsaw 0.7 Roma 1.9 9.6 21 27 Florence 0.1 Palermo 0.1 Benares 0.2 Dublin 0.3 Amsterdam 0.3 Tientsin 0.7 Milano 1.9 Guangzhou 9.6 28 Prague 0.1 Moscow 0.1 Manchester 0.2 Lisbon 0.3 0.3 Shanghai 0.6 1.9 Tientsin 9.5 29 Aleppo 0.1 Milan 0.1 Glasgow 0.2 Cairo 0.3 Brussels 0.3 Newcastle 0.6 1.8 Kinshasa 9.4 30 Tunis 0.1 Madrid 0.1 Madras 0.2 Patna 0.2 Budapest 0.3 St. Louis 0.6 Glasgow 1.8 9.2 Notes: The main sources of the data are Chandler (1987); United Nations (2014). We use the years 1825 and 1900 instead of the years 1800 and 1910 as in the analysis on urbanization rates, because Chandler (1987) only lists the 100 largest cities for the years 1825 and 1900. 9 10 R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21

The absolute size of the largest cities has grown dramatically over time, and in particular after 1950. Beijing, the largest city in 1500, had around 700,000 inhabitants, and even by 1825 had grown to only 1.4 million. There were only 8 cities that were significantly larger than 100,000 inhabitants in 1500, and in 1700 only 20. Even through the 19th century, the largest city sizes remained small by contemporary standards. London in 1875 was 4.2 million people, and 6.2 million by 1900. If London had not grown after 1900, it still would have been among the largest 5 cities in 1950. But by 2010 a population of 6.2 million is smaller than Dacca, Manila, Lagos, and Kinshasa, and would not put a city in the top Fig. 6. Maximum and mean city size, 100–1950. Notes: The main 30 cities in the world. Fig. 6 shows the shift in the scale sources of the data are Chandler (1987); United Nations (2014). The maximal city size is the size of the largest city in the world for of mega-cities over time. It plots the maximum city size different years. The mean city size is the mean size of all cities with at by year from 100 to 1950. Extending the figure to least 300,000 inhabitants for different years. We use 300,000 because further years makes the scale all but useless, demon- it is the lowest population threshold for which we have data in 1950. strating the massive growth of cities since 1950. From Years beyond 1950 are not shown because the scale becomes maximum city sizes of around 500–700,000 between unreadable. 100 and 1700, this accelerates so that maximum city size is over 5 million by 1900 and is over 12 million by 1950. The right-hand axis of Fig. 6 measures the urbanization rates among developing countries seen in average size of a mega-city, defined here as cities over the prior section. 300,000 inhabitants.6 The average is around 500,000 This leads to the second general pattern observed until 1700, and then by 1950 is close to one million from the location of mega-cities over time. For the inhabitants. In 2010, the average size of mega-cities is period running from roughly 1500 until the mid-20th approximately 1.3 million. This confirms that the century, the largest cities in absolute size were typically increase in the absolute size of cities was not only located in the most advanced nations of their time. With driven by the largest cities on earth, but for all cities the onset of the Industrial Revolution the association of across the board. mega-cities with economic development became even The acceleration in the size of mega-cities means that stronger. Looking at Table 1, in 1825 London and Paris absolute growth has risen over time. The fastest change are among the largest five cities in the world, but the top in population ever experienced by London was 93,000 30 cities range across Europe and Asia, with no North new residents per year during the 1890s. New York American cities on the list. As the Industrial Revolution grew by 220,000 new residents per year during the begins to spread, however, by 1850 and 1875 the nature 1920s, and Los Angeles by 248,000 per year during the of the list has changed. By 1875 the top cities are 1950s. In comparison, mega-cities in developing London, Paris, New York, and Berlin. Beijing and countries of the late 20th century are growing at Istanbul are still among the largest 10 cities in the absolute rates unseen in history. During the 2000s, world, but they have barely grown (or actually shrunk in Delhi added 620,000 residents per year, Beijing size) between 1825–1875. As one moves down the list 603,000 per year, and Dacca 445,000. The raw flow in 1875, industrializing cities such as Philadelphia, of people into developing mega-cities is two to three Liverpool, Glasgow, Manchester, Birmingham, and times the maximum flow seen by mega-cities in the Boston have entered the list. past. Absolute growth in the size of mega-cities in By 1900, as mentioned above, the list of largest cities developing countries is an analogue of the higher is heavily concentrated in industrializing nations. Beijing and Istanbul have moved farther down the list, roughly stagnant in size while they are passed by cities such as Vienna, Tokyo, and Manchester. In 1950, the 6 The threshold of 300,000 is used because this is the lowest same pattern holds, with the majority of the largest threshold for which we have city size data in 1950–2010, and for cities located in the industrial countries of Europe and consistency we apply this threshold to previous years. North America. R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21 11

Fig. 7 plots the fraction of mega-cities, defined as 30%, as the rich nations of today industrialized and those larger than 300,000 inhabitants, that are found in experienced rapid urbanization. But following the rush what are currently developing nations over time. To of urbanization in currently rich nations, the developing clarify, we take a contemporary threshold of “develop- world begins to catch up. By 1950, already, the fraction ing” countries (GDP p.c. less than $12,000 in 2000, as of largest cities in developing nations is back to 45%, per the International Monetary Definition), and look and this fraction has climbed ever higher since then. It is backwards in time at their share of the largest cities in currently close to 80%. The share of total population in the world.7 The black solid line shows this fraction, and large cities (the gray line) follows the share of cities as can be seen in 1700 this is about 45%. By 1850 this themselves almost exactly. If we use 1 million as the fraction is roughly the same, indicating that close to half cut-off for a large city, then the post-1900 pattern holds of the largest cities in the world were in what are now up as well. Finally, eliminating China and from considered developing countries. There was no distinct the calculation lowers the share of cities in developing advantage to currently rich nations in the number of nations across time, but we can again see the dip large cities prior to 1850. towards 1900 and then the catch-up by developing However, the share of mega-cities in developing nations afterward. The rise of mega-cities in poor countries drops appreciably by 1900, to only about countries was thus not due to the fast urbanization of China and India in the modern period.8 This matches the pattern seen in urbanization rates in the prior section. Places that industrialized during this period saw the largest absolute growth in city sizes, and also tended to gain in both urbanization rate and measured GDP per capita. From the mid-20th century forward, however, we see the shift towards rapidly growing poor mega-cities in developing countries. Further, there is evidence that city size in the past was a robust indicator of city-level living standards, while that relationship has broken down in more recent years. We do not have a consistent, comprehensive way to measure city-level living standards over time. But by combining several sources we believe we can show how the relationships have changed. For the pre-1910 Fig. 7. Share of large cities in developing vs. developed countries, period, our data are for welfare ratios calculated using 1700–2010. Notes: The main sources of the data are Chandler (1987); Maddison (2008); Bolt et al. (2014); United Nations (2014). nominal wages and price indices for minimal consump- “Number” and “Population” show the relative shares of cities in tion baskets in different cities (Allen, 2007; Allen et al., developing countries in terms of number of cities above 300,000 2011, 2012; Frankema and Van Waijenburg, 2012; inhabitants and total population of these cities, respectively. “Excl. Francis, 2013; Bassino et al., 2014). The 111 observa- China and India” and “N1 Million” show the relative share of cities in tions are at the city-year level, so that for many cities we developing countries in terms of number of cities when excluding China and India or when restricting the sample to cities above have multiple observations over time. We rank all these 1 million inhabitants, respectively. We do not show the year for which observations based on their absolute city size. We then we have fewer than 5 cities (1500 when using 300,000 as the threshold rank all these observations based on their real wage, and or 1500–1875 using one million). “Developing countries” are defined plot the rank of real wages against the rank of city size.9 as those with log per capita GDP lower than 9.42 in 2010 (roughly $12,000), equivalent to the level in Slovakia in that year. Slovakia was the last country to graduate to the category of developed countries before the year 2010 according to International Monetary Fund 8 This pattern is robust to using other thresholds for city size – 500,000, (2009). The gray dashed line represents the year 1910. 2millionand5millioninhabitants– see Web Appendix Fig. A.21. 9 We use ranks rather than logs or levels for two reasons. First, we want a means of looking at the relationship that is comparable to later periods 7 We use per capita income in 2010 rather than per capita income in where we do not have the same welfare ratios. Second, city sizes have a each year to distinguish developed from developing countries. Indeed, skewed distribution and using ranks keeps the relationship from being all countries including today's developed countries were developing driven by outliers. We use all cities that have data on a welfare ratio and countries initially. Using income in 2010 thus allows us to distinguish that are greater than 100,000 inhabitants. This gives us 111 observations. cities in developing countries that industrialized and other developing We can restrict this to 61 cities over 300,000 inhabitants, and the results are countries that did not. similar (see Web Appendix Fig. A.22). 12 R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21

Fig. 8. City living standard rank versus city size rank, historically (pre-1910) and in 2010. Notes: This graph displays the relationship between city living standards and city size for 111 cities of more than 300,000 inhabitants in 2010 and 111 city-year observations of more than 100,000 inhabitants pre-1910 (we use multiple observations for a same city). For each period, we rank the cities by living standards and city size and show the correlation between the two (the linear fit is estimated using as weights the population of each city-year observation). City living standards are proxied by city product indexes in 2000–2010 and welfare ratios for the pre-1910 period. The two sources used to obtain the city product index in 2010 are United Nations Habitat (1998, 2012). The sources used to obtain the welfare ratios for the pre-1910 period (estimated for a “bare bones” consumption basket) are Allen (2007); Allen et al. (2011, 2012), Frankema and Van Waijenburg (2012); Francis (2013), Bassino et al. (2014). We obtain the size of each city from Chandler (1987); United Nations (2014).

Fig. 8 plots the rank of real wages against the rank of reports. For these observations we again rank them by city size for these historical observations, and there is a size, and rank them by city product index, and then plot positive relationship (the dashed line). This is not to say the rank of living standard against the rank of size. In that the mega-cities of industrializing Europe or North Fig. 8 these are also plotted along with a linear fit (the America had high absolute living standards. These solid line). The correlation in 2010 is only half the size, cities were congested and unhealthy. But city size did at 0.28, than in the historical data. The largest seem to indicate something regarding relative living mega-cities today are thus twice less likely to be the standards at the time, and larger cities tended to have richest cities than in the past.10 higher real wages. Likewise, Rome in 300 AD before To highlight the differences over time, several the collapse of Roman Empire was not an exception to individual cities are highlighted in the figure. Amster- the rule, since its ranks in terms of real wages and size dam, London, and New York all have relatively high were relatively close. The correlation in rank between ranked living standards, both before and after 1910. But living standards and city size is approximately 0.6, note that their relative rankings in city size have where a correlation of 1.0 would indicate that city living changed over time. From 1700 to 1900 both Amsterdam standards lined up perfectly with size of city. and London shift to the right, indicating they are larger For the modern period, we do not have real wage relative to other cities in the pre-1910 ear. We do not measures comparable to historical data. However, we have living standard data for New York from 1700, but do have a city product index for a sample of 111 cities its growth in size from 1700 to 1900 is clear. London in of at least 300,000 inhabitants in 2010 (United Nations Habitat, 1998; United Nations Habitat, 2012). The 10 We thus compare 111 cities larger than 300,000 inhabitants in product index is measured using such inputs as capital 2010 with 111 cities larger than 100,000 inhabitants pre-1910. The results are robust to also using a threshold of 300,000 inhabitants for investment, formal versus informal employment, trade, cities in the pre-1910 period. As we have only 61 cities above 300,000 savings, exports and imports, and household income inhabitants pre-1910, we only consider the 61 largest cities in 2010. and consumption. Full details are available in the UN See Web Appendix Fig. A.22. R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21 13

1900 is the largest city in this pre-1910 data. New York Second, from the mid-20th century to the present we in 1900 is also one of the largest cities of the time. All have seen a distinct change in this pattern. Urbanization three are ranked among the richest cities in the world in rates have risen for the very poorest countries, and the 1900. From that point forward, however, they slip down location of mega-cities is shifting towards poor, the rankings in size while maintaining their position in developing countries and away from rich, developed living standards. By 2010 Amsterdam is no longer even ones. The correlation between urbanization and devel- in the top 100 cities by size, London has dropped to opment has been falling demonstrably over the late 20th 25th, and New York to 8th. century when compared to the historical experience. In comparison are a number of currently poor While developing countries in the 20th century have mega-cities. Lagos in 1910, Jakarta in 1900, Delhi in experienced urbanization without growth, this is not 1875, and Kolkata in 1825 are all relatively small, and unique to them, nor does it constitute an anomaly from relatively poor. If we compare them to their contempo- the perspective of historical patterns of urbanization. rary positions, they have all moved up to become among the largest cities of the world in 2010. However, 3. Urbanization and development in theory this has not been associated with a move up in the rankings in living standards. They have grown in How are we to understand the urbanization without absolute size, but this has not caught them up to places growth at all income levels between 1500 and 2010? like Amsterdam, London, and New York in living Further, what explains the fact that this shift occurred standards. Other megacities in the developing world first in the richest countries between 1500 and the such as Dacca and Kinshasa are among the largest cities mid-20th century, and then took hold only in the latter in the world while being also among the poorest. half the 20th century among the poorest countries? The patterns in Fig. 8 are thus in line with the results In this section we review theories of urbanization for urbanization rates seen in the prior section. We have and development to explore the possible answers to urbanization without growth across all cities of the these questions. We first present a simple framework for world, so to speak, with absolute sizes growing thinking about urbanization and development in order everywhere. However, this occurred first in the richest to provide a consistent means of comparing theories. cities, and only later in the 20th century was matched by growth among the poorest. 3.1. Conceptual framework

2.4. Summary of empirical analysis From an accounting perspective we can write

Given our evidence regarding urbanization rates and y wr uwu−wr ; 1 mega-cities, several patterns stand out in the relationship ¼ þ ðÞ ð Þ of urbanization and development. If we compare urbanization and development in 1500 to 2010, there where y is income per capita, wr are rural earnings per has been a distinct level shift in urbanization rates over (rural) worker, wu are urban earnings per (urban) time at all levels of income per capita. On average, worker, and u is the urbanization rate. From this we urbanization rates are about 25–30 percentage points can see that there is a relationship between urbanization higher today than in 1500. Urbanization has diffused and income per capita if wu ≠ wr. Empirically, it may across all levels of development over the last 500 years, be the case that wu N wr at a given time or in a given and one could term this shift urbanization without country, and hence increased urbanization would lead to growth. higher average income per capita. But this shift took place in two distinct phases. First, The urbanization rate, u, will depend on the relative from the period 1500 to the mid-20th century, utility of people in the two areas. Denote rural utility urbanization rose rapidly for the richest countries, Ur(wr, Xr) and urban utility Uu(wu, Xu), where Xr and while remaining stagnant for the poorest. The location Xu capture all the ancillary characteristics of these areas of the largest cities was concentrated in the richest areas (e.g. amenities or mortality rates). Utility in both sectors of the world in this period. In particular it is during the is increasing in earnings and the level of X. If early 20th century, overlapping with the widespread Uu(wu, Xu) N Ur(wr, Xr) then there is migration from industrialization of Europe and North America, when rural to urban areas, and u is increasing. In the case the correlation of urbanization and income per capita where Uu(wu, Xu) b Ur(wr, Xr), then u would be becomes the most pronounced. decreasing. The economy will tend towards an 14 R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21

equilibrium where Uu(wu, Xu)=Ur(wr, Xr), but this 3.2.2. Industrial revolutions transition could take years or decades. This approach describes how a rise in industrial productivity attracts underemployed labor from agri- 3.2. Urbanization with growth culture into the industrial sector (Lewis, 1954; Hansen and Prescott, 2002; Lucas, 2004; Alvarez-Cuadrado The experience of the classical Industrial Revolution and Poschke, 2011), which is presumed to be more has informed much of the literature on urbanization and concentrated in urban areas than is agricultural development, which often uses it as an example of how production.12 This approach either assumes that there urbanization and structural transformation interact to is no food problem or that there is some other means produce higher levels of GDP per capita. These theories of meeting subsistence food requirements. There could reflect urbanization with growth, in that urbanization is be surplus labor in the agricultural sector, as in the seen as either a consequence of productivity growth that dual economy model of Lewis (1954),ortheindustrial causes structural change into predominantly urban could be preceded by an agricultural revolution, as in sectors, or a cause of productivity growth in the economy Asia where the occurred early due to agglomeration effects. To simplify greatly, these (Evenson and Gollin, 2003). Alternatively, an indus- theories give explanations for why the slope of the trial revolution could directly facilitate the moderni- relationship between urbanization and GDP per capita is zation of agriculture through better agricultural positive, but do not necessarily explain why the level of intermediate inputs (Yi and Zhang, 2011). Lastly, a urbanization rose over time at all levels of GDP per capita. country could export manufactured goods (or services) We discuss first theories where urbanization is a to import food (Matsuyama, 1992; Teigner, 2011; Yi consequence (agricultural, industrial, and resource revo- and Zhang, 2011). Regardless of the exact mechanism, lutions) of economic growth before turning to urbaniza- in these models an industrial revolution draws labor tion as a cause of economic growth. into urban areas in response to a productivity improvement. 3.2.1. Agricultural revolutions In poor countries, large fractions of land and labor are devoted to producing food for subsistence needs (Schultz, 3.2.3. Resource revolutions 1953; Gollin et al., 2002, 2007). This “food problem” Revenue windfalls from resource extraction – oil, prevents the reallocation of productive resources to other gold, and diamonds – may also drive urbanization. sectors. An agricultural revolution provides an increase in Gollin et al. (2013), Jedwab (2013), and Cavalcanti et food productivity that reduces the food problem and al. (2014) show that resource revenues provide releases labor for the modern and/or urban sector additional income, as well as having a significant effect 13 (Matsuyama, 1992; Caselli and Wilbur John Coleman, on urbanization rates. The resource revenues are spent 2001; Gollin et al., 2002, 2007; Nunn and Qian, 2011; disproportionately on urban goods and services, either Michaels et al., 2012; Motamed et al., 2014). This because of income effects similar to an agricultural structural change could be the consequence of income revolution or the concentration of revenues in the hands effects: non-homothetic preferences and rising incomes of an urban elite. These resource revolutions may not mean a reallocation of expenditure shares towards non- have the same long-run effects on growth as food goods (Engel's law). Or it could be the consequence agricultural or industrial revolutions as they tend to of price effects: assuming a low elasticity of substitution produce “consumption cities” where the workforce is across consumption goods, any increase in the productiv- ity of the food sector leads to a decrease in its employment share. In either case the non-food sectors are predomi- 12 nantly urban, which is consistent with data on residence In the longer run, developed countries do eventually deindustri- 11 alize, and specialize in tradable services such as finance and business and sector of employment (Gollin et al., 2013). services, thus producing a rise and fall of manufacturing (Herrendorf et al., 2011; Buera and Kaboski, 2012). More recently, a country may 11 Bustos et al. (2013) suggest that the factor bias of agricultural directly specialize in tradable services (Ghani and Kharas, 2010; technology matters for whether labor is released from that sector or Gollin et al., 2013), as exemplified by , Macao, or not. Existing theories typically presume that productivity is labor- and Hyderabad in India. Other examples include merchant cities of the saving and so leads to urbanization. Further, with trade in food 15th–17th century (Acemoglu et al., 2005). anything that lower the world price could lead to urbanization even in 13 More generally, any rent (e.g., international aid, remittances or countries that do not experience a productivity improvement (Glaeser, illicit trafficking in arms and drug) could produce urbanization due to 2013; Gollin et al., 2013). similar forces. R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21 15 concentrated in non-tradable services as opposed to or indirectly (through the low income elasticity for food), tradable industrial goods or services. and ceteris paribus, raises Uu(wu, Xu) N Ur(wr, Xr). Urbanization increases as workers migrate from rural to 3.2.4. Agglomeration effects urban areas in search of higher urban earnings. If Unlike the prior two theories, the economic geogra- agglomeration effects are present, then wu is itself a phy literature suggests that causation runs from function of u (or absolute city size) and this creates a urbanization to productivity in both developed countries positive feedback where increased urbanization leads to (Rosenthal and Strange, 2004; Henderson, 2005; higher wages and even more in-migration to urban areas. Glaeser and Gottlieb, 2009; Combes et al., 2012; Thus, in each of these theories countries “climb up” the Duranton and Puga, 2013) and developing countries line relating urbanization and GDP per capita. (Overman and Venables, 2005; Henderson, 2010; Further exaggerating the relationship between ur- Felkner and Townsend, 2011; Duranton, 2014). Cities banization and development was, perversely, the low allow countries to save on the transportation costs of living standards in cities of this period. These were goods, people and ideas (see Glaeser and Gottlieb “killer cities” (Williamson, 1990; Clark and Cummins, (2009) for a thorough survey of the literature). First, by 2009; Voigtlaender and Voth, 2009; Voigtlaender and agglomerating thousands or millions of people in a few Voth, 2013; Christiaensen et al., 2013; Jedwab and locations, cities facilitate the distribution of goods. Vollrath, 2015) with high densities, industrial smoke, Firms are closer to suppliers. They are close to polluted water sources and unhygienic practices con- consumers, and vice versa. Second, cities create thick tributing to high urban mortality. As a result, industrial labor markets and are centers of human capital cities could not grow without massive inflows of accumulation. In particular, returns-to-scale imply that migrants willing to move to unhealthy urban environ- universities and hospitals are located in large cities. ments. Rural migration was thus the main driver of Third, if there are agglomeration economies within urban growth (Williamson, 1990; Kim, 2007; sectors (localization economies) or across sectors Christiaensen et al., 2013; Jedwab and Vollrath, (urbanization economies), firms directly benefit from 2015). The poor conditions in cities would imply that being located close to firms of the same or different in terms of the non-pecuniary measures Xu b Xr. Hence industries. As long as these agglomeration effects equilibrium would require wu N wr to ensure that dominate congestion effects, cities increase productivity Uu(wu, Xu)=Ur(wr, Xr) in equilibrium. With wu N wr, and wages in the long run. Empirical evidence suggest any increase in urbanization would have mechanically that these mechanisms are also at play in developing raised measured income per capita, as noted in Eq. (1), countries (Henderson et al., 2001; Lall et al., 2004; da even if this did not involve any improvement in welfare. Mata et al., 2007; Deichmann et al., 2008; Henderson, Furthermore, in a Malthusian setting, this urbanization 2010) and many have argued that urbanization may would have led to higher equilibrium wu and wr by promote growth in these countries (Duranton, 2008; raising mortality rates through warfare and plague World Bank, 2009; Venables, 2010; McKinsey,, 2011). (Voigtlaender and Voth, 2013). Retrospectively, the models described above appear The same theories are somewhat successful in to describe well the experience of European (and explaining the urbanization processes of Latin America Neo-European) countries over the period from the late and Asia in the 20th century. The agricultural revolution 19th century to 1950 (Bairoch and Goertz, 1986; and the industrial revolution first diffused to the Bairoch, 1988; Williamson, 1990; Kim and Margo, Neo-European countries of Latin America in the late 2004; Kim, 2006; Allen, 2009). The agricultural 19th century. , and Uruguay were revolution created a food surplus, which helped among the most urbanized countries in 1950 (their establish and consolidate urban networks across Eu- urbanization rates were respectively 65%, 58% and rope. Whether the agricultural revolution directly 78%). The revolutions then diffused to much of the rest preceded the original Industrial Revolution, or was of the Latin American region in the 20th century necessary for it at all, is beyond the scope of this paper. (Kingsley and Casis, 1946; Bairoch, 1988; Machicado What we know is that cities in Europe and the et al., 2012). The urbanization rate of Latin American Neo-Europe grew as industrial centers, as seen in the and the Caribbean region as a whole increased from pattern of mega-city growth in this period. about 20% in 1900 to 40% in 1950 and 80% in 2010. In our framework, the source of urbanization is More recently, agricultural and industrial revolutions fundamentally a change in productivity. This raises wu have diffused to , and to a lesser extent, relative to wr either directly (as in an industrial revolution) South-East Asia (Bairoch, 1988; Bosworth and Collins, 16 R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21

2008; Brandt et al., 2008). The urbanization rate of Asia by specific kinds of (urban) technological progress has increased from about 10% in 1900 to 15% in 1950 (Bairoch, 1988; Glaeser, 2011) over the last one and 45% in 2010. The following countries have hundred years. Given a walking speed of 5 km per developed and urbanized (or are in the process of doing hour, the size of cities was traditionally limited to about so) over the course of the 20th century: , South 20 sq km. Obviously, cities are able to grow larger Korea, , , , , when people could use horse carriages. This all changed and China. with the transportation (and architectural) revolutions of Over this same period, the revenues from resource the 19th and 20th centuries. First, railways (trains, extraction have led to significant increases in urbanization tramways and metro lines) and roads (cars, buses, and income per capita in many countries (Gollin et al., motorbikes and bicycles) have reduced transportation 2013). Oil producers of the Middle-East (e.g. Kuwait, costs for goods and people. Second, cities have also ) are obvious examples, but countries such become denser, after the invention of cheap steel, and as Mongolia, Malaysia, and have also urban- the elevator, which allowed the construction of taller ized rapidly while exporting resources. In Sub-Saharan buildings. Whether cities are becoming denser (building Africa, the wealthiest and most urbanized countries are up) or sprawling (building out), it is relatively easier to those with significant resource exports, such as , be “urban” in the 21st century now that these Botswana, , and . Resource extrac- technologies have diffused across countries. tion was historically a source of urbanization in the US South (cotton and tobacco), the Caribbean islands (sugar 3.3.2. Urban amenities and preferences and cotton), and South America (silver, copper, and more In the Rosen–Roback model (Rosen, 1979; Roback, recently oil). 1982, 1988), rents and wages are higher in cities with a better quality of life, because people directly value it in 3.3. Urbanization without growth their utility function. In other words, they are willing to “pay” for a better quality of life (by accepting a lower The theories just reviewed link urbanization to net income). This utility differential is often explained productivity improvements explicitly, but do not by urban amenities or the natural features of some cities provide a way of understanding how urbanization may (the diversity of restaurants and entertainment venues, rise even in the absence of those productivity improve- the proximity to a river or the coast, etc.). Cities also ments. Our empirical work showed that urbanization have advantages in consumption (Glaeser et al., 2001; has risen for all countries over time, not just those that Rappaport et al., 2003). For example, Glaeser et al. were blessed with rich resource deposits or that (2001) show how the demand for living in cities has experienced industrialization. risen over time in the U.S. A historical amenity of cities The urbanization without growth that we document- may have been their role as safe havens from violent ed at every level of income per capita is not necessarily conflict (Glaeser and Shapiro, 2002; Dincecco and a bad thing, even though it is not linked to productivity Gaetano Onorato, 2013) or from persecution by feudal explicitly. Positive sources of urbanization without lords (Pirenne, 1925, 1936). growth include changes in urban amenities due to the diffusion of urban technologies that limit congestion and disease, or changes in preferences towards urban 3.3.3. Urban bias living. Of course, there are also negative sources that Urbanization without growth is often attributed to induce higher urban through distortions to the urban bias (Lipton, 1977; Bates, 1981; Bairoch, prices or labor allocations without necessarily improv- 1988), i.e. urban-biased policies such as agricultural ing urban amenities. We discuss first positive sources of overtaxation, public employment in the manufacturing urbanization without growth, followed by negative and service sectors (e.g., the government sector), and sources. food price subsidies for urban residents. Such distor- tions artificially increase the urban–rural income gap, 14 3.3.1. Urban technologies which fuels migration. If public employment is As noted previously, the largest cities of today's concentrated in the largest city, the urban bias may developing world are much larger than the largest cities 14 Symmetrically, rural-biased policies, or migration restrictions, of the previous centuries. The list of largest cities in the should “artificially” reduce the urbanization rate. Au and Vernon world in 2010 is dominated by developing country Henderson (2006) show that a large fraction of cities in China are cities. The growth of these mega-cities was accelerated undersized due to strong migration restrictions. R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21 17 also lead to urban primacy (Ades et al., 1995; Davis and Increased urban utility would draw migrants from rural Vernon Henderson, 2003). Urbanization here does not areas and keep existing urban residents from leaving. come from income growth but from sectoral distortions Thus urbanization could increase without any change in 15 (a reallocation of income across areas, given a constant earnings wu or wr. income level). From a historical perspective, this type of Negative theories of urbanization without growth urban bias is perhaps evident in Rome, Tenochtitlan, work somewhat differently. These policies may use Beijing, and Byzantium as they grew as administrative subsidies and/or taxes to raise wu relative to wr, centers of large empires without necessarily being inducing people to move from rural to urban areas. As associated with sustained growth in income per capita. this represents a distortion to the efficient allocation, this would lower the level of both wu and wr. 3.3.4. Rural poverty Alternatively, if urban amenities Xu are increased by Rural poverty (whether it is due to agricultural extracting rural resources, then this will induce a shift of overtaxation or another factor), land pressure (due to population towards urban areas, but at the cost of lower demographic growth, given land is inelastically supplied) rural wages relative to urban areas. In both cases, and man-made or natural disasters (e.g., wars that destroy urbanization is associated with lower earnings per villages or climate change) constitute rural push factors worker. feeding rural exodus (World Bank, 2000; Barrios et al., In the period from 1500 to 1950, there appear to be 2006; Poelhekke, 2011; Henderson et al., 2013a). These some clear features of positive urbanization without factors result in a lower rural wage, the urban–rural growth occurring. Without taking a stand on causality, income gap increases, and poor migrants flock to the industrialization and improvements in urban technolo- cities. These migrants find employment in the gies occur within a similar time period. Otis' elevator low-productivity urban sectors if urban labor markets arrives in 1852. The first skyscraper (the 10-story Home have a limited absorption capacity. Insurance Building in Chicago) was built in 1885, and the first all-steel framed skyscraper arrived in 1889. For 3.3.5. Internal urban growth public sanitation, John Snow demonstrated the correla- Historical experience suggests that urbanization is tion of cholera and water sources in London in 1854, primarily associated with in-migration. However, that and that city began regulating water supply standards in no longer appears to be the case in many developing 1855. Chlorination of drinking water to eliminate germs countries. With the onset of the mortality transition after begins in the 1890s. The availability of clean drinking World War II, urban mortality rates plummeted in most water and sewage service meant that 20th century cities nations. This has led to rapid natural increase within were no longer “killers”. cities (Rogers, 1978; Keyfitz, 1980; Rogers and Transportation technologies like the steamboat and Williamson, 1982; Potts, 1995; Christiaensen et al., railroad, which arrived in the early 19th century, made 2013; Jedwab and Vollrath, 2015) as urban fertility supplying cities cheaper, allowing them to grow. remained high while mortality fell to low levels. Intra-city rail networks arrived in the mid-19th century Christiaensen et al. (2013) call these “mushroom cities” (e.g. The Metropolitan Railway began service in 1863 for these cities, which grow through “urban push” as in London). Automobiles arrived in the late 19th opposed to the commonly used “rural push” (such as century, and by 1911 the first highway, the Long Island from an agricultural revolution) and “urban pull” (as Motor Parkway, had opened. Taken all together, these from an industrial revolution). This implies that cities technologies made cities more attractive places to live, grow not as a result of income growth, but due to a shift allowing urbanization rates to rise without necessarily in demographic behavior. The rapid natural increase of meaning that productivity increased. Urbanization the urban population drives up absolute city sizes and without growth was a feature of the industrializing tends to raise the urbanization rate, although there is not world, even if there as also significant urbanization with necessarily any change in development levels. growth occurring at the same time. In terms of our framework, these various theories For the post-1950 era, some combination of these suggest several ways in which urbanization and positive and negative forces seems to be a plausible way development need not be closely related. Consider the positive theories of urbanization without growth — the 15 Rural poverty would operate differently, by lowering Xr and diffusion of urban technologies and increasing urban making rural areas inherently less attractive. This would also drive amenities. In these theories Uu(wu, Xu) may rise due to more population into urban areas without an appreciable change in factors influencing Xu, the inherent utility in . development levels. 18 R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21 of thinking about the shift up in the level of This urbanization without growth occurred it two urbanization at low levels of GDP per capita. The distinct phases, however. From 1500 to roughly 1950 or mortality transition (Stolnitz, 1955; Davis, 1956; 1960, urbanization rates increased dramatically in the Preston, 1975; Acemoglu and Johnson, 2007) lowered richest countries in the world, and the growth of the urban mortality substantially in developing countries in largest cities was concentrated in those rich nations. this period and allowed for both “urban push” But at that point the relationship has changed. demographic growth as well as making cities more Urbanization without growth in the late 20th century attractive locations for workers. Developing countries has occurred in the poorest countries of the developing were able avoid the “killer cities” of the rich world's world. The absolute size of cities has increased rapidly history (Fogel, 2004), which create urbanization in the poorest countries during the late 20th century, and without growth by leaving more people alive in cities city size is becoming less indicative of city living and making them more attractive destinations. standards than in the past. However, there also appears to have been significant These historical comparisons provide several lessons urbanization without growth for negative reasons occur- for studying urbanization and growth. First, they caution ring in these developing countries. The poor mega-cities us not to presume that urbanization and industrialization of the late 20th do not take advantage of innovations in or development are synonymous. Urbanization has city technology (e.g. public transportation, skyscrapers, occurred throughout history without necessarily being sanitation) to an extent consistent with their size. Urban associated with manufacturing expansion or with rapid bias (Lipton, 1977; Bates, 1981; Bairoch, 1988; Ades et economic development, as during an industrial revolu- al., 1995) appears to be very strong with developing tion. This urbanization without growth may be good (e.g. countries in the late 20th century, particularly in Africa. due to lower urban mortality rates) or bad (e.g. Rural poverty, driven either by increasing density, rent-seeking in capital cities) for welfare. Second, climate change, or wars, has driven more residents to theories that are built to explain the positive correlation mega-cities, again representing a negative source of of urbanization and income per capita, typically involving urbanization without growth. sectoral shifts following a productivity improvement, are Urbanization without growth is not limited to the not sufficient to understand all of the long-run shifts in period from 1500–2010, of course. One could argue urbanization rates seen in the world. that the cities of Ancient Greece, the Roman Empire, the Aztecs, the European Renaissance, Ottoman Em- Appendix A. Data sources pire, Moghul India, China during the Qing dynasty all displayed urbanization without growth even if they may Most of our analysis is at the country level, and so we have been examples of commercial “efflorescences” require country-level data on urbanization rates and (Braudel, 1973). These booms in urbanization (Rome development levels. For development, we use log GDP may have reached 1 million residents at its peak) could per capita (PPP, constant 1990 dollars) from Bolt et al. have been for positive reasons of better amenities such (2014), who update Maddison (2008). We base our as regular water supplies, or represented urban bias sample of the 159 countries available from these sources, associated with empire-building. Regardless, it seems which account for 99% of the world population in 2010. clear that urbanization without growth is not a process To the information on GDP per capita we add unique to the late 20th century. observations on urbanization rates for a selection of years running from 1500 to 2010. We use primarily Bairoch (1988); Acemoglu et al. (2002); Malanima and 4. Discussion and conclusion Volckart (2007); United Nations (2014); Jedwab and Moradi (Forthcoming) to obtain the urbanization rates, By putting the relationship of urbanization and supplemented by observations from several additional development in historical perspective, it is possible to sources. The availability of urbanization and GDP data see that urbanization without growth documented in the leaves us with 1319 total observations from the 159 late 20th century is actually a continuation of a process countries. As one might expect, we have far more data running back to 1500. Over the last five centuries from later rather than earlier years. The distribution of urbanization rates have risen at every level of GDP per observations across time is for 1500 (24 observations), capita, so that urbanization rates are now about 25–30 1700 (18), 1800 (25), 1850 (23), 1870 (46), 1910 (70), percentage points higher at all income levels than they 1950 (159), 1960 (159), 1970 (159), 1980 (159), 1990 were in 1500. (159), 2000 (159), and 2010 (159). R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21 19

We also examine evidence on city size and city living Bairoch, Paul, Goertz, Gary, 1986. Factors of urbanisation in the standards, although this information is less widely nineteenth century developed countries: a descriptive and econometric analysis. Urban Stud. 23 (4), 285–305. available. For city size, we use Chandler (1987); United Barrios, Salvador, Bertinelli, Luisito, Strobl, Eric, 2006. Climatic Nations (2014),andlookatthetop100citiesbysizeorall change and rural–urban migration: the case of Sub-Saharan cities larger than 300,000 inhabitants in several select Africa. J. Urban Econ. 60 (3), 357–371. years. We do not have comparable measures of city living Bassino, Jean-Pascal, Fukao, Kyoji, Takashima, Masanori, 2014. A standards across all time periods. For the pre-1910 period First Escape from Poverty in Late Medieval Japan: Evidence From Real Wages in Kyoto (1360–1860). we use welfare ratios based on a “bare bones” consumption Bates, Robert, 1981. Markets and States in Tropical Africa: the basket as reported in Allen (2007, 2011, 2012); Frankema Political Basis of Agricultural Policies. University of California and Van Waijenburg, 2012; Francis (2013); Bassino et al. Press, Berkeley. (2014) and this yields 111 observations of city size and Bolt, Jutta, Luiten, Jan, van Zanden, 2014. The Maddison Project: living standards in the pre-1910 period (which includes collaborative research on historical national accounts. Econ. Hist. Rev. 67 (3), 627–651. cities observed in morethanoneyear)forcitieslargerthan Bosworth, Barry, Collins, Susan M., 2008. Accounting for growth: 100,000 inhabitants. For 2010, we use city product indexes comparing China and India. J. Econ. Perspect. 22 (1), 45–66. from (United Nations Habitat, 1998, 2012). The data is Brandt, Loren, Hsieh, Chang-Tai, Zhu, Xiaodong, 2008. Growth and available for 157 cities but we focus on the 111 largest structural transformation in China. In: Brandt, L., Rawski, T.G. cities to have a consistent sample size for both periods. (Eds.), China's Great Economic Transformation. Cambridge University Press, Cambridge, pp. 683–728. Braudel, Fernand, 1973. Capitalism and Material Life. Harpercollins, Appendix B. Supplementary data London. Buera, Francisco J., Kaboski, Joseph P., 2012. The rise of the service economy. Am. Econ. Rev. 102 (6), 2540–2569. Supplementary data to this article can be found Bustos, Paula, Caprettini, Bruno, Ponticelli, Jacopo, 2013. Barcelona online at http://dx.doi.org/10.1016/j.eeh.2015.09.002. Graduate School of Economics Working Papers 736. Caselli, Francesco, Wilbur John Coleman, I.I., 2001. The U.S. structural transformation and regional convergence: a reinterpre- References tation. J. Polit. Econ. 109 (3), 584–616. Cavalcanti, Tiago, Da Mata, Daniel, Toscani, Frederik, 2014. Acemoglu, Daron, Johnson, Simon, 2007. Disease and development: Winning the oil lottery: the impact of natural resource extraction the effect of life expectancy on economic growth. J. Polit. Econ. on growth. European Regional Science Association ERSA 115 (6), 925–985. Conference Papers ersa14p109. Acemoglu, Daron, Johnson, Simon, Robinson, James A., 2002. Chandler, Tertius, 1987. Four Thousand Years of Urban Growth: an Reversal of fortune: geography and institutions in the making of Historical Census. The Edwin Mellen Press. the modern world income distribution. Q. J. Econ. 117 (4), Christiaensen, Luc, Marina Gindelsky, and Remi Jedwab. 2013. 1231–1294. “Rural Push, Urban Pull or. Urban Push? New Historical Evidence Acemoglu, Daron, Johnson, Simon, Robinson, James A., 2005. The from 40 Developing Countries.” Unpublished manuscript, George rise of Europe: Atlantic trade, institutional change, and economic Washington University, Department of Economics. growth. Am. Econ. Rev. 95 (3), 546–579. Clark, Gregory, 2009. Contours of the world economy, 1–2030 AD: Ades, Alberto F., Edward, L., Glaeser, 1995. Trade and circuses: essays in macro-economic history. By Angus Maddison. Oxford: explaining urban giants. Q. J. Econ. 110 (1), 195–227. Oxford University Press, 2007. J. Econ. Hist. 69 (04), Allen, Robert C., 2007. How Prosperous Were the Romans? Evidence 1156–1161. From Diocletian's Price Edict (301 AD). Clark, Gregory, Cummins, Neil, 2009. Urbanization, mortality, and Allen, Robert, 2009. The British Industrial Revolution in Global fertility in Malthusian England. Am. Econ. Rev. 99 (2), 242–247. Perspective. Cambridge University Press, Cambridge. Combes, Pierre-Philippe, Duranton, Gilles, Gobillon, Laurent, Puga, Allen, Robert C., Bassino, Jean-Pascal, Ma, Debin, Moll-Murata, Diego, Roux, Sebastien, 2012. The productivity advantages of Christine, Zanden, Jan Luiten Van, 2011. Wages, prices, and large cities: distinguishing agglomeration from firm selection. living standards in China, 1738–1925: in comparison with Econometrica 80 (6), 2543–2594. Europe, Japan, and India. Econ. Hist. Rev. 64 (s1), 8–38. da Mata, D., Deichmann, U., Henderson, J.V., Lall, S.V., Wang, H.G., Allen, Robert C., Murphy, Tommy E., Schneider, Eric B., 2012. The 2007. Determinants of city growth in . J. Urban Econ. 62 colonial origins of the divergence in the Americas: a labor market (2), 252–272. approach. J. Econ. Hist. 72 (04), 863–894. Davis, Kingsley, 1956. The amazing decline of mortality in Alvarez-Cuadrado, Francisco, Poschke, Markus, 2011. Structural underdeveloped areas. Am. Econ. Rev. Pap. Proc. 46, 305–318. change out of agriculture: labor push versus labor pull. Am. Davis, Mike, 2007. Planet of Slums. Verso Books, London. Econ. J. 3 (3), 127–158. Davis, James C., Vernon Henderson, J., 2003. Evidence on the Au, Chun-Chung, Vernon Henderson, J., 2006. Are Chinese cities too political economy of the urbanization process. J. Urban Econ. 53 small? Rev. Econ. Stud. 73 (3), 549–576. (1), 98–125. Bairoch, Paul, 1988. Cities and Economic Development: From the De, Long, Bradford, J., Shleifer, Andrei, 1993. Princes and merchants: Dawn of History to the Present. The University of Chicago Press, European city growth before the industrial revolution. J. Law Chicago. Econ. 36 (2), 671–702. 20 R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21

Deaton, Angus, Heston, Alan, 2010. Understanding PPPs and PPP- Hansen, Gary D., Prescott, Edward C., 2002. Malthus to Solow. Am. based national accounts. Am. Econ. J. 2 (4), 1–35. Econ. Rev. 92 (4), 1205–1217. Deichmann, Uwe, Lall, Somik V., Redding, Stephen J., Venables, Hausman, Jerry, and Zhong Liu. 2014. “Mobile Phones in Developing Anthony J., 2008. Industrial location in developing countries. Countries.” Unpublished manuscript. World Bank Res. Obs. 23 (2), 219–246. Henderson, J. Vernon, 2005. Urbanization and growth. In: Aghion, P., Dincecco, Mark, Gaetano Onorato, Massimiliano, 2013. Military Durlauf, S. (Eds.), Handbook of Economic Growth 1. Elsevier, conflict and the rise of urban Europe. IMT Institute for Advanced pp. 1543–1591 (Chapter 24). Studies Lucca Working Papers 7/2013. Henderson, J. Vernon, 2010. Cities and development. J. Reg. Sci. 50 Dittmar, Jeremiah E., 2011. Information technology and economic (1), 515–540. change: the impact of the printing press. Q. J. Econ. 126 (3), Henderson, Vernon, Lee, Todd, Lee, Yung Joon, 2001. Scale 1133–1172. externalities in Korea. J. Urban Econ. 49 (3), 479–504. Duranton, Gilles, 2008. Viewpoint: from cities to productivity and Henderson, Vernon, Adam Storeygard, and Uwe Deichmann. 2013. growth in developing countries. Can. J. Econ. 41 (3), 689–736. “Has Climate Change Promoted Urbanization in Sub-Saharan Duranton, Gilles, 2014. Growing through cities in developing Africa?” Unpublished manuscript, Department of Economics, countries. The World Bank Policy Research Working Paper Series Brown University 6818. Henderson, Vernon, Mark Roberts, and Adam Storeygard. 2013. “Is Duranton, Gilles, and Diego Puga. 2013. “The Growth of Cities.” Urbanization in Sub-Saharan Africa Different?” Unpublished Unpublished manuscript, The University of Pennsylvania Wharton manuscript, Department of Economics, Brown Univerisity. School, Real Estate Department. Herrendorf, Berthold, Rogerson, Richard, Valentinyi, Akos, 2011. Evenson, Robert E., Gollin, Douglas, 2003. Assessing the impact of Growth and structural transformation. Draft Prepared for the the green revolution, 1960 to 2000. Science 300 (5620), 758–762. Handbook of Economic Growth. Fay, Marianne, Opal, Charlotte, 2000. Urbanization without growth: a Hoselitz, Bert F., 1953. The role of cities in the economic growth of not-so-uncommon phenomenon. The World Bank Policy Re- underdeveloped countries. J. Polit. Econ. 61, 195. search Working Paper Series 2412. Hoselitz, Bert F., 1955. Generative and parasitic cities. Econ. Dev. Felkner, John S., Townsend, Robert M., 2011. The geographic Cult. Chang. 3 (3), 278–294. concentration of enterprise in developing countries. Q. J. Econ. International Monetary Fund, 2009. World Economic Outlook. IMF, 126 (4), 2005–2061. Washington, DC. Fogel, Robert W., 2004. The Escape From Hunger and Premature Jedwab, Remi. 2013. “Urbanization without Industrialization: Evi- Death, 1700–2100. Cambridge University Press. dence from Consumption Cities in Africa.” Unpublished manu- Francis Jr., Joseph, 2013. The Terms of Trade and the Rise of script, Department of Economics, George Washington University Argentina in the Long Nineteenth Century. Jedwab, Remi, Moradi, Alexander, 2015g. The permanent effects of Frankema, Ewout, Van Waijenburg, Marlous, 2012. Structural transportation revolutions in poor countries: evidence from Africa. impediments to African growth? New evidence from real wages Rev. Econ. Stat. (Forthcoming). in British Africa, 1880–1965. J. Econ. Hist. 72 (04), 895–926. Jedwab, Remi, and Dietrich Vollrath. 2015. “The Mortality Transi- Ghani, Ejaz, Kharas, Homi, 2010. The Service Revolution. World tion, Malthusian Dynamics, and the Rise of Poor Mega-cities.” Bank Econ. Premise 14, 1–5. Unpublished manuscript Glaeser, Edward, 2011. Triumph of the City: How Our Greatest Jerven, Morten, 2014. Economic Growth and Measurement Invention Makes Us Richer, Smarter, Greener, Healthier, and Reconsidered in Botswana, , , and Zambia, Happier. Penguin Press, London, UK. 1965–1995. Oxford University Press, OUP Catalogue. Glaeser, Edward L., 2013. A World of Cities: the Causes and Keyfitz, Nathan, 1980. Do cities grow by natural increase or by Consequences of Urbanization in Poorer Countries. National migration? Geogr. Anal. 12, 142–156 (url =). Bureau of Economic Research, Inc NBER Working Papers 19745. Kim, Sukkoo, 2006. Division of labor and the rise of cities: evidence Glaeser, Edward L., Gottlieb, Joshua D., 2009. The wealth of cities: from U.S. industrialization, 1850–1880. Working Papers 12246. agglomeration economies and spatial equilibrium in the United National Bureau of Economic Research, Inc NBER. States. J. Econ. Lit. 47 (4), 983–1028. Kim, Sukkoo, 2007. Immigration, industrial revolution and urban Glaeser, Edward L., Gyourko, Joseph, 2005. Urban decline and growth in the , 1820–1920: factor endowments, durable housing. J. Polit. Econ. 113 (2), 345–375. technology and geography. Working Papers 12900. National Glaeser, Edward L., Shapiro, Jesse M., 2002. Cities and warfare: the Bureau of Economic Research, Inc NBER. impact of terrorism on urban form. J. Urban Econ. 51 (2), Kim, Sukkoo, Margo, Robert A., 2004. Historical perspectives on 205–224. U.S. economic geography. In: Henderson, J.V., Thisse, J.F. (Eds.), Glaeser, Edward L., Kolko, Jed, Saiz, Albert, 2001. Consumer city. Handbook of Regional and . Handbook of J. Econ. Geogr. 1 (1), 27–50. Regional and Urban Economics 4. Elsevier, pp. 2981–3019 Gollin, Douglas, Parente, Stephen L., Rogerson, Richard, 2002. The (Chapter 66). role of agriculture in development. Am. Econ. Rev. 92 (2), Kingsley, Davis, Casis, Ana, 1946. Urbanization in Latin America. 160–164. Milbank Mem. Fund Q. 24 (2), 186–207. Gollin, Douglas, Parente, Stephen L., Rogerson, Richard, 2007. The Kuznets, Simon, 1968. Modern Economic Growth: Rate, Structure, food problem and the evolution of international income levels. and Spread. Yale University Press, New Haven, CT. J. Monet. Econ. 54 (4), 1230–1255. Lall, Somik V., Shalizi, Zmarak, Deichmann, Uwe, 2004. Agglom- Gollin, Douglas, Remi Jedwab, and Dietrich Vollrath. 2013. eration economies and productivity in Indian industry. J. Dev. “Urbanization with and without Industrialization.” Unpublished Econ. 73 (2), 643–673. manuscript, Oxford University Department of International Lewis, Arthur, 1954. Economic development with unlimited supplies Development. of labour. Manch. Sch. 22 (2), 139–191. R. Jedwab, D. Vollrath / Explorations in Economic History 58 (2015) 1–21 21

Lipton, Michael, 1977. Why Poor People Stay Poor: Urban Bias in Rogers, Andrei, 1978. Migration, urbanization, resources & World Development. Harvard University Press, Cambridge. development, and discussions. In: McMains, H., Wilcox, L. Lucas, Robert E., 2004. Life earnings and rural–urban migration. (Eds.), Alternatives for Growth: the Engineering and Economics J. Polit. Econ. 112 (S1), S29–S59. of Natural Resources Development. Elsevier, pp. 147–238 Machicado, Carlos, Rioja, Felix, Saravia, Antonio, 2012. Productivity, (Chapter 9). structural change, and Latin American development. Working Rogers, Andrei, Williamson, Jeffrey Gale, 1982. Migration, urbani- Paper Series 03/2012. Institute for Advanced Development zation, and third world development: an overview. Econ. Dev. Studies Development Research. Cult. Chang. 30 (3), 463–482. Maddison, Angus, 2008. Statistics on World Population, GDP and Per Rosen, Sherwin, 1979. Wage-based indexes of urban quality of life. Capita GDP, 1–2008 AD. GGDC, Groningen. In: Miezkowski, Peter, Straszheim, Mahlon (Eds.), Current Issues Malanima, Paolo, Volckart, Oliver, 2007. Urbanisation 1700–1780. in Urban Economics. Johns Hopkins University Press, Chapter Prepared for the Third RTN/CEPR Summer Symposium pp. 74–104. in London: an Economic History of Modern Europe. Rosenthal, Stuart S., Strange, William C., 2004. Evidence on the Matsuyama, Kiminori, 1992. Agricultural productivity, comparative nature and sources of agglomeration economies. In: Henderson, advantage, and economic growth. J. Econ. Theory 58 (2), J.V., Thisse, J.F. (Eds.), Handbook of Regional and Urban 317–334. Economics 4. Elsevier, pp. 2119–2171 (Chapter 49). McKinsey, 2011. Lions on the Move: the Progress and Potential of Schultz, Theodore W., 1953. The Economic Organization of African Economies. Mc Kinsey Global Institute, London. Agriculture. McGraw-Hill, New York. Michaels, Guy, Rauch, Ferdinand, Redding, Stephen J., 2012. Stolnitz, George J., 1955. A century of international mortality trends: Urbanization and structural transformation. Q. J. Econ. 127 (2), I. Popul. Stud. 9, 24–55. 535–586. Teigner, Marc. 2011. “The Role of Trade in Structural Transforma- Motamed, Mesbah, Florax, Raymond, Masters, William, 2014. tion.” Unpublished manuscript, University of Chicago, Depart- Agriculture, transportation and the timing of urbanization: global ment of Economics. analysis at the grid cell level. J. Econ. Growth 19 (3), 339–368. United Nations, 2014. World Urbanization Prospects: the 2014 Nordhaus, William D., 1996. Do real-output and real-wage measures Revision. United Nations, Department of Economic and Social capture reality? The history of lighting suggests not Chapters The Affairs, New York. Economics of New Goods. NBER, National Bureau of Economic United Nations Habitat, 1998. Global Urban Indicators Database. UN- Research, Inc., pp. 27–70. Habitat, , Kenya. Nunn, Nathan, Qian, Nancy, 2011. The potato's contribution to United Nations Habitat, 2012. State of the World's Cities Report population and urbanization: evidence from an historical experiment. 2012/2013: Prosperity of Cities. UN-Habitat, Nairobi, Kenya. Q. J. Econ. 126 (2). Venables, Anthony J., 2010. Economic geography and African Overman, Henry, Venables, Anthony J., 2005. Cities in the development. Pap. Reg. Sci. 89 (3), 469–483. developing world. Centre for Economic Performance, LSE CEP Voigtlaender, Nico, Voth, Hans-Joachim, 2009. Malthusian Discussion Papers dp0695. dynamism and the rise of Europe: make war, not love. Am. Pirenne, Henri, 1925. Medieval Cities, Their Origins, and the Revival Econ. Rev. 99 (2), 248–254. of Trade. Princeton, NJ. Voigtlaender, Nico, Voth, Hans-Joachim, 2013. The three horsemen Pirenne, Henri, 1936. Economic and Social History of Medieval of riches: plague, war and urbanization in early modern Europe Europe. Harcourt Brace, London. Forthcoming in the Rev. Econ. Stud. 80 (2), 774–811. Poelhekke, Steven, 2011. Urban growth and uninsured rural risk: Williamson, Jeffrey, 1990. Coping With City Growth During the British booming towns in bust times. J. Dev. Econ. 96 (2), 461–475. Industrial Revolution. Cambridge University Press, Cambridge. Potts, Deborah, 1995. Shall we go home? Increasing urban poverty in World Bank, 2000. World Development Report 2000: Entering the African cities and migration processes. Geogr. J. 161 (3), 21st Century. World Bank Publications, Washington DC. 245–264. World Bank, 2009. World Development Report 2009: Reshaping Potts, Deborah, 2012. Challenging the myths of urban dynamics in Economic Geography. World Bank Publications, Washington DC. Sub-Saharan Africa: the evidence from Nigeria. World Dev. 40 Yi, Kei-Mu., Zhang, Jing, 2011. Structural change in an open (7), 1382–1393. economy. Fed. Reserv. Bank Minneapolis Staff Rep. 456. Preston, Samuel H., 1975. The changing relation between mortality Xu, Yi, van Leeuwen, Bas, van Zanden, Jan Luiten, 2015. and level of economic development. Popul. Stud. 29, 231–248. Urbanization in China, ca. 1100–1900. Working Paper Series Rappaport, Jordan, Jeffrey, D., Sachs, 2003. The United States as a 63. Centre for Global Economic History CGEH. coastal nation. J. Econ. Growth 8 (1), 5–46. Young, Alwyn, 2012. The African growth miracle. J. Polit. Econ. 120 Roback, Jennifer, 1982. Wages, rents, and the quality of life. J. Polit. (4), 696–739. Econ. 90 (6), 1257–1278. Roback, Jennifer, 1988. Wages, rents, and amenities: differences among workers and regions. Econ. Inq. 26 (1), 23–41.